PCI 6115 malfunctioning when plugged into a PCIE to PCI slot

[miliescu]

Hi, I developped a DAQ system containing 6 PCI 6115 high memory option(64MB), 2 PCI RIO 7113R and one VME2PCI bridge, which pilots and acquires digital and analog data from an experiment. The data on the PCI 6115 AI come in bursts, sampled at 10MS/sec on all 24 channels, the whole process being guided by the two 3Mgate FPGAs. The system works fine with deterministic data transfer rates of about 700MB/s, on an industrial PC equipped with a P4 processor board and an IEI expansion backplane, which provides the necessary number of PCI slots.

Now the problem: the slow side of the system (external, seen trough the VME bridge) imposes some low frequency (40 Hz) stops and requires the reset of the whole PCI setup. This creates some dead time (7%), mainly due to the 6 ADC tasks to be stopped and restarted (altough only the final part of the 4 steps procedure is implied, while verify-reserve-comit are called only once, at the launch). The dead-time was acceptable until recently, when a rate increase on the VME by a factor of 4 brought a too large data loss. The solution found was a better CPU/memory configuration (C2Q, which reloads the task twice faster) but imposed also a backplane change. The new one (PICMG 1.3, 19 slot, PXE-19S by IEI which, unfortunately, seems to be the only solution) has 4 "real" PCI slots plus other 12 provided trough 3 PCIE to PCI bridges (PEX8111). 

The ADCs mounted in the bridge slots behave erratically (mix the data trough channels, don't pass the self-test, etc.) while those mounted in the normal PCIs work well (same PC). The CPU board is a PCIE-Q350, where the southbridge provides 4PCI, 6PCIEx1 (4 of which became our 12 PCI) and 1 PCIEx16. 

The error reported by NIMAX during self-test is -200020. My concern is the DMA operation of the 6115 boards trough the new bridges. The backplane is marked PICMG 1.3 so in principle the slots should fulfill the whole PCI standards but it appears there are some issues with our ADCs. During program execution no apparent problem arises, one just finds out when analysing that data from "normal" PCI ADCs is fine while that from PCIE bridge shows strong channel mix, maybe also out-of clock conversion.. The system extension and complexity (spares included), as well as its connection with non-NI devices does not allow us to drastically change the standard (go to PXI, changing the versions, etc.).  Any help to better understand the reason of bad data transfer from 6115 to PC memory passing trough PCIE would be greatly appreciated.

Thanks,

Mihai

[Hudson]

Hi Mihai

It sounds like your boards are functioning properly in the standard PCI slots, but when going through the PCIe to PCI bridge (PEX 8111) you are experiencing erratic behavior. I found that PCI boards should be compatible with PICMG 1.3 but NI can't guarantee that 3rd party bridges are going to be up to spec, you might want to contact the manufacturer for help on this one.

Best Regards

HUDSON

[mikewogden]

Interested if you had any more progress with this problem.

We had a similar problem, in that when the DAQ cards are inserted  into the PCI-X slots then occasionally the DAQMX virtual address gets corrupt.

New address numbers are allocated to these cards whilst graying out the original allocated virtual addresses.

The application then fails as it get no loger find the physical cards.

Mike